GSA DATA REPOSITORY

GSA DATA REPOSITORY 2016167 SUPPLEMENTAL INFORMATION APPENDIX DR1 Barenblatt (1962) shows that the stress intensity factor K 0 for a fluid-filled crack of length l with a uniform fluid pressure P f,0 within a homogeneous isotropic lithostatic stress field of magnitude lith is given by K 0 = P (l/2), (DR1-1) where P = P f,0 lith is the net pressure. See also Barenblatt (2014, chapter 5) for an abbreviated analysis. A tensile crack expands in a critical fashion if the stress intensity factor K 0 is equal to or exceeds a critical value K c (Griffith s criterion), that is, K 0 K c. (DR1-2) Following Barenblatt (1962), Aki et al. (1977) consider the following situation: the original crack is fully filled with fluid. During crack expansion the fluid front cannot keep up with the moving crack tip, creating a fluid lag zone, that is, an area at the crack tip which is free of fluids. The crack advances by a distance l, and the fluid front is assumed to be stationary for simplicity. The fluid-free zone is at zero pressure. Crack extension then creates a cohesive force, or negative stress intensity K 1, equal to K 1-0.9 lith ( l/2), (DR1-3) because of the zero fluid pressure within the fluid-free zone. After expansion, the stress intensity K new,aki at the moved crack tip becomes K new,aki = (P f,0 lith ) (l/2 + l/2) - 0.9 lith ( l/2) = K 0 + (P f,0 lith ) ( l/2) - 0.9 lith ( l/2). (DR1-4) This follows from equations (DR1-1) and (DR1-3) if the new crack length l+ l is taken into account. If crack extension occurs for K 0 K c, then Aki et al. (1977) argue that the presence of a fluid lag zone will halt crack extension since K new,aki < K c for a reasonable net pressure P and lithostatic stresses at say more than 500m depth (around 10MPa). Their analysis implies that renewed crack growth then only happens once the fluids have reached the crack tip or the fluid

pressure P f inside the crack is increased, thus countering the negative stress intensity due to the fluid lag zone. Aki et al. (1977) argue that the above analysis implies that only continuous fracture growth is possible, since any crack extension l, even an infinitesimally small one, leads to halting. In this scenario, the fluid speed becomes equal to the rupture velocity, a common assumption in mathematical models for hydraulic fracture propagation (Adachi et al., 2007). However, Griffith s criterion, equation DR1-2, only predicts when tensile failure occurs. It does not determine the fracture extension length l. Likewise, the Mohr-Coulomb failure criterion for shearing only predicts if failure is likely, but not the rupture length. Dynamic fracture models are required to predict parameters such as rupture (fracture) length, area and displacement (that is, slip and/or aperture). For instance, the seismic moment M 0 for shear failure is given by M 0 = (16/7) s a 3, (DR1-5) where s is the static stress drop and a is the fracture radius (Eshelby, 1957; Keylis-Borok, 1959; Walter and Brune, 1993). Equation DR1-5 holds for a Poisson s body (that is, equal Lamé parameters or ). The static stress drop s is defined as the difference between shear stress across the fault before and after the event. A stress drop causes a reduction in driving stress. Shear failure halts when the shear stress becomes less than the frictional stress and cohesive forces. Equation DR1-5 shows that the seismic moment M 0 and thus the size of an event is proportional to the stress drop s, which in turn is determined by the material and stress heterogeneity. In a similar fashion, it can be shown that the seismic moment M 0 for a tensile event is equal to M 0 = 22 P s a 3, (DR1-6) where P s is the static fluid pressure drop and fracture radius a= l/2. Again a Poisson s body is assumed. Equation DR1-6 is derived using the following relationships: (1) the nonzero moment tensor components for a tensile crack in the horizontal plane and opening along the vertical axis, are M 11 = M 22 = AD and M 33 =( +2 )AD, with A= a 2, the surface area, and D is average displacement (aperture); (2) the moment M 0 is defined as M 0 = (M ik M ik /2) with summation over all indices i and k of the full moment tensor (Julian et al., 1998); and (3) the average aperture D

= 2a P s /( ) (Sneddon, 1951; Walter and Brune, 1993; Eaton et al., 2014b). Tensile cracks oriented in other directions yield the same result. Note that in analogy with equation DR1-5, variable P s represents a static stress drop, whereas Walter and Brune (1993) and Eaton et al. (2014b) assume it is equal to the effective net pressure. If the argument of Aki et al. (1977) that any crack extension, including an infinitesimally small one, leads to halting is applied to the Mohr-Coulomb failure criterion, it predicts that only slow slip is possible but no abrupt shear failure (and thus no felt seismicity, equation DR1-5), since it assumes a medium with homogeneous material properties (e.g., no asperities) and homogeneous stresses. Likewise, stress and material heterogeneity ensures episodic tensile crack growth where the seismic moment M 0 of individual events is given by equation DR1-6. For instance, in our numerical simulations the presence of natural fracture sets and intact rock bridges ensures stress and material heterogeneity. It is important to note that variations in material properties generally tend to cause stress heterogeneity [e.g., Roche and Van der Baan (2015)]. The above analysis thus supports episodic instead of continuous crack propagation for hydraulic fracturing treatments which are often around 2km depth, targeting heterogeneous rocks such as shales. Furthermore, the negative stress intensity, equation (DR1-3), also leads to partial closure of the crack near its tip, creating a process reminiscent of hand clapping. Continuous and possibly even accelerated fracture growth is however possible at shallow depths since the lithostatic stress lith becomes then small, reducing the effect of the closing negative stress intensity, equation (DR1-3). Secor (1969) invokes a related but different reasoning to argue for episodic crack growth. In his model, crack extension leads to additional crack volume, temporarily reducing the fluid pressure P f at the crack tip, thus lowering the stress intensity factor. Specifically the new fluid pressure P f,new changes from the original fluid pressure P f,o as P f,new = P f,o c -1 V/V, (DR1-7) with c the fluid compressibility coefficient and V/V the relative increase in crack volume (not necessarily uniform across the entire crack). For a perfect gas c -1 P f,o. Therefore, a 10% increase in volume leads to a 10% reduction in fluid pressure near the crack tip. For water, c -1 = 1.96x10 9 Pa, and a 1% increase in volume corresponds to a 19.6MPa reduction in fluid pressure. This is

substantial if we consider for instance a reservoir around 2km depth with 10MPa net pressure (P f,o = 60MPa, lith = 50MPa). The original stress intensity factor K 0 prior to crack extension is again given by equation (DR1-1). After crack extension the new stress intensity factor K new,secor then becomes K new,secor = (P f,o (1 c -1 V/V) lith ) (l/2 + l/2) = K 0 + (P f,o lith ) ( l/2) - (c -1 V/V) (l/2 + l/2). (DR1-8) If we assume again that crack extension happens when K 0 K c, then the reduction in fluid pressure at the crack tip is likely to cause halting of the crack propagation since K new,secor < K c due to the negative third term in equation (DR1-8). Crack propagation continues when new fluids have reverted the fluid pressure within the additional volume V to the original fluid pressure P f,o, and the third term becomes close to zero again. In addition, if the new fluid pressure P f,new becomes less than the lithostatic stress lith then the net pressure P at the crack tip becomes negative, causing again partial crack closure, as indicated by equation (DR1-1). Equation (DR1-8) implies there is a trade-off between increased stress intensity due to the increased crack length and reduced stress intensity due to volume increase and thus drop in fluid pressure at the crack tip. In practice, the third term in equation (DR1-8), reduced stress intensity due to a local pressure drop, is likely to be somewhat tempered since (1) the height of most hydraulic fractures is substantially smaller than their length, whereas the above analysis assumes a crack that is infinite in one direction, and (2) the pressure drop only occurs at the crack tip, not across the entire crack length. A constricted height limits the total moment (that is, force times crack length) exerted on the crack tip, thus reducing the effective crack length. Likewise, the effective crack length is reduced since only the portion where the pressure drop occurs should be taken into account. Nonetheless, considering some representative numbers such as a reservoir around 2km depth with 10MPa net pressure (P f,o = 60MPa, lith = 50MPa) and a 5m effective fracture that extends by 1 meter (l=5m, l=1m) with merely a 1% increase in volume ( V/V= 0.01) demonstrates the third term is generally dominant. The importance of the second term increases with decreasing depth, that is, decreasing lithostatic stresses,, pointing again at the possibility of continuous fracture growth at shallow depths. Aki et al. (1977) and Secor (1969) invoke different causal mechanisms, yet both analyses predict episodic instead of continuous crack growth for anthropogenic hydraulic fracturing treatments at common depths due to reduced fluid pressures at the crack tip. Secor s (1969)

model does not involve a fluid lag zone, that is, a fluid-free zone at the crack tip. A fluid lag zone has been seen in laboratory investigations (Medlin and Masse, 1984; Groeneboom et al., 2003). The size of the fluid lag zone is however unclear and subject to theoretical investigations (Garagash and Detournay, 2000); yet its existence will encourage temporal halting of fracture propagation due to a local pressure drop with associated partial closing. Continuous fracture growth is often inferred in case of natural hydraulic fracturing (Bahat and Engelder, 1984; Savalli and Engelder, 2005). The above analysis points to several conditions under which this might occur at critical stress intensities, including very high net pressures P, shallow depths and thus small lithostatic stresses lith, fracturing due to inflow of gas which is highly compressible, growth into mechanically weak zones or into areas with low effective stresses (that is, due to elevated pore pressures or reduced lithostatic stresses, for instance, because of upward growth), and finally a reduction in critical stress intensity K c during fracture propagation, equations (DR1-2), (DR1-4) and (DR1-8). See also Lacazette and Engelder (1992) who discuss the effect of fluid infiltration into the hydraulic fracture, compressibility of the fluid inside the hydraulic fracture, and the effect of fluid flow velocities (in particular, Secor s 1969 model) on episodic versus continuous fracture growth at critical stress intensities. A reduction in critical stress intensity K c during tensile fracture propagation would imply the existence of a larger, static and a smaller, dynamic critical stress intensity for tensile failure, analogous to static and dynamic shear friction coefficients governing shear failure (Marone, 1998; Rubinstein et al., 2004). Tests however show that the dynamic tensile strength of rocks (measured at high strain rates) is larger than the quasi-static strength (Rubin and Ahrens, 1991), likely due to stress redistribution within samples and crack arrests because of increasing microcrack densities with increasing strain rates (Cho et al., 2003). The mathematical analysis thus confirms the numerical results of stick-split behavior and repeated opening and closing (hand clapping) near anthropogenic hydraulic fracture tips.

APPENDIX DR2 Far-field body wave spectra radiated from a small circular crack can be represented using a variant of the Brune source model (Walter and Brune, 1993). Eaton et al. (2014b) analyzed microseismic data from a hydraulic fracturing treatment targeting the Montney formation in British Columbia, Canada (Eaton et al., 2013). Their Figure 13 shows one microseismic event displaying spectral notching. We analyzed microseismic events that occurred during a fracturing treatment into a tight-sand formation in Alberta, Canada at ~1860m depth (Eaton et al., 2014a). The microseismic events were recorded using a 12-level triaxial geophone array installed in a vertical monitor well, in a depth range from 1606m to 1835 m. The procedure used to process each microseismic dataset is similar to Oye and Roth (2003) and includes the following steps: 1) application of an instrument-response correction to convert recorded signals into units of displacement; 2) rotation of 3-component traces into raycentered co-ordinates, thereby isolating P, S fast and S slow signals; 3) stacking of P, S fast and S slow waveforms by beam forming using an iterative cross-correlation approach with static shifts and polarity checking, similar to the method of De Meersman et al. (2009); 4) calculation of P and S displacement spectra by computing the Fourier transform of the recorded waveforms within a 250ms window centered on the body-wave arrival, where anisotropy of the radiated S wave spectrum is accounted for using the vector sum of the S fast and S slow signals. The noise spectrum is extracted from a 250ms pre-event window. The best-fitting Brune spectrum and clapping spectrum are obtained using a least-squares fitting approach (Eaton et al., 2014b), which is parameterised based on quality factor (Q) for P and S waves, corner frequency, low-frequency plateau amplitude and time interval between opening and closing. The S/P amplitude ratio is measured using the fitted low-frequency plateau amplitudes for P and S waves. Following Eaton et al. (2014b), the tensile crack radius a is estimated using log 10 (a) = 3.05 - (1/6)log 10 (22) + M w /2 (1/3)log 10 ( P s ), (DR2-1) where P s is the static fluid pressure drop, equation DR1-6, which for simplicity we take to be equal to the net pressure P (fluid pressure minus minimum principal stress, equation DR1-1). The latter is the pressure that is required to keep the fracture open. Equation DR2-1 is derived by taking the logarithm of equation DR1-6 and combining this with M w = (2/3)log 10 (M0)-6.1.

The tensile aperture E can be obtained from equation (3) in Walter and Brune (1993), yielding E = 3a P /( ), (DR2-2) with the shear modulus of the intact rock. REFERENCES CITED (in APPENDICES DR1 and DR2) Adachi, J., Siebrits, E., Peirce, A., Desroches, J., 2007, Computer simulation of hydraulic fractures: Int. J. Rock Mech. Min. Sci. Abstr., 44, 739 757. Aki, K., Fehler, M., and Das, S., 1977, Source mechanism of volcanic tremors: Fluid-driven crack model and their application to the 1963 Kilauea eruption. J. Volcanol. Geotherm. Res., 2 (3), 259-287 Bahat D. and Engelder, T., 1984, Surface morphology and cross-fold joints of the Appalachian Plateau, New York and Pennsylvania, Tectonophysics, 104, 299-313. Barenblatt, G.I., 1962. The mathematical theory of equilibrium cracks in brittle fracture. Adv. Appl. Mech., 7: 56-129. Barenblatt, G.I., 2014: Flow, Deformation and Fracture, Cambridge Univ Press. Cho, S.H., Ogata, Y. and Kaneko, K., 2003, Strain-rate dependency of the dynamic tensile strength of rock, International Journal of Rock Mechanics and Mining Sciences, 40(5), 763 777. De Meersman, K., Kendall, J-M. and Van der Baan, M., 2009, The 1998 Valhall microseismicity: An integrated study of relocated sources, seismic multiplets and S-wave splitting. Geophysics 74(5), B183-B195 Eaton, D., Van der Baan, M., Tary, J-B., Birkelo, B., Spriggs, N., Cutten, S. and Pike, K., 2013, Broadband microseismic observations from a Montney hydraulic fracture treatment, northeastern B.C., Canada. CSEG Recorder, 38(3), 45-53 Eaton D., Caffagni, E., Rafiq, A., Van der Baan, M., Roche, V. and Matthews, L., 2014a, Passive seismic monitoring of a tight-sand reservoir during hydraulic-fracture treatment, flowback and production. Paper 1929223, Unconventional Resources Technology Conference, Denver, CO.

Eaton, D. W., Van der Baan, M., Birkelo, B. and Tary, J-B., 2014b, Scaling relations and spectral characteristics of tensile microseisms: Evidence for opening/closing cracks during hydraulic fracturing. Geophys. J. Int., 196(3), 1844-1857 Eshelby, I.D., 1957, The determination of the elastic field of an ellipsoidal inclusion and related problems, Proc. R. Soc. London, Ser. A, 241, 376-396. Garagash D. and Detournay E., 2000, The tip region of a fluid-driven fracture in an elastic medium. ASME Journal of Applied Mechanics, 67, 183-192. Groenenboom, J., van Dam, D.B., and de Pater, C.J., 2001, Time-lapse ultrasonic measurements of laboratory hydraulic-fracture growth: Tip behavior and width profile. SPE Journal, 6(1), 14-24, SPE-68882. Julian, B.R., Miller, A.D. & Foulger, G.R., 1998. Non-double-couple earthquakes. I. Theory, Rev. Geophys., 36, 525 549. Keylis-Borok, V.I., 1959, On the estimation of the displacement in an earthquake source and of source dimensions, Ann. Geofis. 12, 205-214. Lacazette, A. and Engelder, T., 1992, Fluid-driven cyclic propagation of a joint in the Ithaca Siltstone, Appalachian Basin, New York: p. 297-323 in B. Evans and T.-F. Wong (editors): Fault Mechanics and Transport Properties of Rocks; a festschrift in honor of W. F. Brace: Academic Press, San Diego. Marone, C., 1998, Laboratory-derived friction laws and their application to seismic faulting, Annual Review of Earth and Planetary Sciences, 26, 643-696. Medlin, W.L. and Masse L., 1984, Laboratory experiments in fracture propagation. SPE Journal, 24, 3, 256-268, SPE-10377. Oye, V. and Roth, M., 2003, Automated seismic event location for hydrocarbon reservoirs. Computers & Geosciences 29(7), 851-863 Roche V. and Van der Baan M. (2015) The role of lithological layering and pore pressure on fluid-induced microseismicity. Journal of Geophysical Research: Solid Earth, 120 (2), 923-943. Rubinstein, S.M., Cohen, G., and Fineberg, G., 2004, Detachment fronts and the onset of dynamic friction, Nature, 430, 1005-1009. Rubin, A.M. and Ahrens, T. J, 1991, Dynamic tensile-failure-induced velocity deficits in rock, Geophysical Research Letters, 18(2), 219-222. Savalli, L. and Engelder T., 2005,

Mechanisms controlling rupture shape during subcritical growth of joints in layered rocks, Geological Society of America Bulletin, 117 (3-4), 436-449. Secor, D. T., 1969, Mechanics of natural extension fracturing at depth in the earth s crust: Geological Survey of Canada Paper 68-52, 3-48. Sneddon, I.N., 1951. Fourier Transforms, McGraw-Hill. Walter, W. R. and Brune, J.N., 1993, Spectra of seismic radiation from a tensile crack. J. Geophys. Res., 98, B3, 4449-4459

Table DR1: Parametric inputs used in the numerical model for stick-split verification, based on in situ conditions in an underground mine in New South Wales, Australia, after Preisig et al. (2015). Natural fracture network properties Dip direction [degrees] Dip angle [degrees] Spacing [m] Family 1 (only natural fractures) 090 81 2.4 Family 2 (only natural fractures) 270 63 3.6 Family 3 (natural and incipient fractures) 270 15 4.2 Persistence of fractures: fully or variable (see Fig. 3) Rock properties Fractures properties Young modulus E [Pa] 60 10 9 Incipient Natural Poisson ratio ν [-] 0.25 Normal stiffness k n [Pa/m] 1.3 10 11 1.3 10 11 Density ρ [kg/m 3 ] 2700 Shear stiffness [Pa/m] 1.3 10 10 1.3 10 10 Tensile strength T [Pa] -0.5 10 6 0.0 Cohesion C [Pa] 1.0 10 6 0.0 Fluid properties and constants Frictional angle Φ [ ] 30 45 Viscosity μ [Pa s] 0.001 Aperture at zero effective 2.0 10-5 2.0 10-5 Density ρ w [kg/m 3 ] 1000 normal stress a 0 [m] Bulk modulus K w [Pa] 0.1 10 9 Residual aperture a res [m] 4.0 10-6 4.0 10-6 Gravity g [m/s 2 ] 9.81 Dilation angle Ψ [ ] 20 20 In-situ stress state Stress σ [MPa] at depth: 1415 [m] σ 1 = 73 σ 3 = 42 Orientation Horizontal E-W Vertical